Community Research and Development Information Service - CORDIS

Final Report Summary - HYPERIMAGE (Hybrid PET-MR system for concurrent ultra-sensitive imaging)

The HYPERIMAGE project has developed a new MR-compatible detector technology for simultaneous positron emission tomography (PET) and magnetic resonance (MR) imaging. As targeted, the project succeeded in applying this new detector technology in a preclinical PET/MR scanner for a human 3 Tesla (3T) MRI and a whole body PET/MR test system. Both systems were designed, built and tested within the timeframe of the European Union (EU) Seventh Framework Programme (FP7) project of 3.5 years. The three main scientific areas of HYPERIMAGE were distributed over three teams:

- system detector hardware development (WP1 and WP2)
- correction and visualisation software (WP3)
- characterisation and application for cardiovascular diseases (CVD) and oncology (WP4 and WP5).

In WP1, a new MR-compatible detector stack was developed for preclinical and clinical simultaneous PET/MR. Here, the silicon photomultiplier (SiPM) technology was further advanced in terms of stability and performance (low variation of breakdown voltages, and coincidence resolution time, CRT). We finally succeeded in showing SiPMs a CRT of with < 200 ps (FWHM). The ASIC development was executed by partner UH, which also succeeded in developing MR-compatible ASICs in three generations, that have low power consumption (< 30 mW/ channel) and a timing resolution of 50 ps (bin). Furthermore, jointly with Philips and FBK, UH designed the world's first integrated MR-compatible detector stack, consisting of an SIPM-tile, an ASIC-tile and a FPGA-tile, which allows sub nanosecond timing measurements under MR operations.

In WP2, partner Philips has developed a preclinical PET/MR scanner for a human 3T system and a clinical whole body test system. For the first system, a new data acquisition and control environment was created for this new type of digital detectors. Special gamma-transparent radio frequency (RF) coils supporting simultaneous PET/MR were developed for the preclinical and clinical systems. The interference investigation with both systems using the new MR-compatible PET modules finally shows no significant performance degradation.

In WP3, novel attenuation correction techniques using dedicated MR sequences and segmentation technologies for three and four classes were successfully developed and published by the partner IBBT, KCL and Philips. Furthermore, a simulation environment was developed by KCL to create simulated dynamic PET/MR data out of a dynamic MR scan. Different PET reconstruction schemes using the motion fields have been investigated with this tool. HYPERIMAGE also developed the dedicated MR sequences to extract motion vector fields with high accuracy in 2D and 3D.

In WP4, partner KCL and Philips have jointly executed characterisation of this newly developed preclinical system. Characterisation of PET showed transversal image resolution of better than 1.6 mm, which reflects the targeted resolution of this system. This also has not changed under simultaneous acquisition.

CNIC used its pre-clinical department and famous MR expertise to prepare for pre-clinical tests focusing on cardiology make use of plaque imaging in rabbits. The animal PET/MR system provided by WP2 will provide the model to study myocardial infarction, in order to test the accuracy for the simultaneous assessment of myocardial metabolism, perfusion, scar and wall motion.

In WP5, partners UKE and NKI developed a 1-stop-shop protocol for MR investigations and successfully proved the concept of PET guided biopsy. In developing a simultaneous PET/MR system we have faced many technical challenges, causing delays in the project. We therefore could not execute all intended clinical validation tasks within the scope and timing of HYPERIMAGE. Validation activities will be executed with the partners KCL, UKE, and CNIC, after termination of the project. HYPERIMAGE ended on 30 September 2011 as a very successful project and with 19 papers and 30 conference records. Furthermore, 29 of the 32 targeted deliverables and all targeted milestones have been met.

Project context and objectives:

With the leading partners of various areas (from silicon detector technology over advancing methods to preclinical and clinical research), the HYPERIMAGE project aimed to extend the European research in this and adjacent areas to a leading position of Europe. On the long range, it is the aim to improve the European healthcare system, by enabling novel investigations, and monitoring and treatment of diseases. HYPERIMAGE aims to expand the initially presented exemplary applications of concurrent PET/MR towards diseases affecting any part of the human body. The following project objectives were successfully archived:

WP1: Development of MR-compatible detector technology with high time resolution for ToF-PET
WP2: Development of hybrid PET/MR test systems for preclinical and clinical PET/MR
WP3: Development of 4D PET/MR motion, attenuation, and functional data acquisition techniques
WP4+5: Initial image performance tests and validation in preclinical studies in cancer and CVD.

WP1 focused on the development of MR-compatible PET detector stacks and modules, which are the building blocks of all PET/MR systems planned in this project. These modules were upgraded with last submissions of SiPMs and TAC/ADC ASICs during the entire project span. Results already achieved in 2008 ensured stable electrical and mechanical interfaces between the modules and the PET backbone, to facilitate simultaneous work on WP1 and WP2. Thus, the project focused in WP1 on the improvement of the SiPMs, the ASIC, and the interface boards. Partners FBK and UH were mainly responsible for this WP.

WP2 focuses on building preclinical and clinical PET/MR test systems. Important system aspects are addressed here: adaptation of the Tx/Rx animal and WB coil, shielding and electromagnetic decoupling, as well as the design of an MR compatible high throughput PET pre-processing electronics. Emphasis will be put on investigating and reducing the crosstalk between the systems to allow simultaneous image acquisition. Philips will be mainly responsible for this task, having years of experience and leading products in high-end PET and MR imaging domain.

Animal PET/MR test insert

The animal-PET/MR system was designed as an insert into an existing 3T MR Achieva system and could thus be used in different clinical environments. The PET modules are axially arranged in a ring with a diameter suitable for mice and rabbit imaging. The PCBs inside these modules have a specially developed layout to reduce the interaction of the gradient field of the MRI with the SiPMs and the ADC/TDC ASIC. The PET-detector and RF-coil are separated via a dedicated RF shield to prevent RF leakage into the PET modules. The RF-shield was specially designed to minimise the RF coupling to all supply cables. Power feeding of the modules was done in with special cables to reduce the static stray field and the vibrations given by the Lorentz forces inside the B0. Electromagnetic simulations have been used to minimise the interaction. The designs of the RF coils have been done using light material which made them almost transparent for the photons.

Whole-body ToF-PET/MR test system

The purpose of the whole-body-PET/MR test system was to built clinical demonstrator with reduced PET gantry based on a modified 3T Achieva MR to investigate the interference and to demonstrate for the first time simultaneous ToF-PET/MR.

The PET-detector ring represented by a two opposing PET modules was located centrally inside the magnet, surrounded by a split gradient coil. The volume of the PET detector modules was reduced to fit into a small 20 cm gap. The RF body coil was located centrally and was specially designed to be transparent for the gamma-photons. A special RF-shield was used to suppress the leakage of the transmit RF-pulse into the PET detector volume. Under EPI acquisition, a timing (CRT) of 449ps (FWHM) was measured which is to our knowledge the best TOF CRT that has ever reported under simultaneous PET/MR. Also, during this acquisition the singles and coincidence rate remains constant for all applied MR scans. PET gains and energy resolution remains constant during all different scans.

WP3 developed the required special image processing tools and MR sequences that will allow the acquisition of functional data and 4D anatomical information in order to generate 4D attenuation maps. Motion model have been developed using by 1D and 2D MRI image navigators for motion field extraction. As the 1D navigator is very fast, this navigator scheme allows interleaving with additional MR sequence. Optimum MR-navigator position was evaluated using single and pairs of navigators formed from volunteer MRI data. The results suggest that two navigator models are more accurate, but only in the presence of significant cycle to cycle breathing motion variability. For the 2D navigators, free-form non-rigid respiratory motion correction in the thorax based on principal component analysis was used. The 2D image navigators provide enough information to drive models and to differentiate between different breathing patterns. Validation of MR data from 10 volunteers demonstrated improvements of up to 40.5 % over other reported motion modeling approaches. A PET-MR simulation package was developed for accurate simulation of simultaneously acquired dynamic PET data using anatomic and dynamic information from real MR acquisitions. The invented approach is up to 7000x faster compared to conventional Monte Carlo simulation (GATE). This tool will be distributed as part of the open-source STIR package, and example datasets will be made available to the community. Different motion compensation schemes have been implemented and compared.

WP4: After initial functional tests at partner Philips in Aachen, an intensive phase of system characterisation at KCL in London was executed. Main purpose of this was to demonstrate the usefulness of the preclinical PET/MR system for imaging of small animals and to evaluate the impact of the workflow of the prototype system. This activity also included significant effort on debugging, normalisation and calibration of the new PET system. The consortium decided to execute initial planned animal experience beyond the official end of the project. Simultaneous PET-MR experiments are very complex involving collaborators from several disciplines. Studies need to be planned well in advance so that resources are not wasted, and so that animal models are not prepared and then not used. We have therefore been very cautious in scheduling experiments until we are confident all aspects of the study will run reliably. We believe we are now at the point and results from preclinical studies should start to appear over the coming months. The outcome is summaries in the first simultaneous structural phantom that show an transversal resolution of about 1.6 mm, which is close to the predicted resolution of 1.5 mm.

The pre-clinical tests focusing on cardiology make use of plaque imaging in rabbits were postponed after the official end data of the project. The animal PET/MR system provided by WP2 will provide the model to study myocardial infarction, in order to test the accuracy for the simultaneous assessment of myocardial metabolism, perfusion, scar and wall motion. The WP is mainly driven by CNIC having a huge pre-clinical department and famous MR expertise.

WP5: To validate the use of a concurrent PET/MR system in diagnostic oncology, we will start with preclinical validation studies, using animal mouse models of human cancers. Mouse xenograft models bearing a range of human tumours of different origin (melanoma, pancreas and colon), all stably expressing the hNIS gene will be established. Cells were genetically manipulated in such way, that hNIS is permanently overexpressed on those cells.

As image guided biopsy in mice prove to be quite difficult this task will be performed in phantoms with the addition of multimodal marker seeds that can be detected in both PET and MRI. Initial validation experiments with these seeds have nearly been finished. Initial dynamic PET and 'one-stop-shop' protocols are available. The PET protocol performs a series of individual scans. First a blank scan, followed by a number of short scans (to monitor the initial bolus perfusion) and followed by a number of 10 min scans to depict the final tumour uptake. A draft version of a MRI 'one-stop-shop' protocol also exists, but further testing and optimisation is still required.

Sixty-five patients with stage II/III breast cancer scheduled to be treated with NAC were investigated with MRI and PET. Scans were acquired prior and during treatment. At histopathology 32 patients showed complete response after NAC and 33 residual disease. The area under the ROC curve was 0.86 for MRI alone, and 0.94 in combination with PET, yielding 84% sensitivity at 95 % specificity. The accuracy was largest for triple-negative tumours at both imaging modalities, but MRI and PET were complementary in the ER+/HER2- and in the HER2+ subgroups.

In extension of the efforts, we also applied the developed navigation set-up in a patient. We initially focused our clinical proof of concept on laparoscopic navigation towards the tracer deposits in the prostate. It was possible to use the navigation system during a robot assisted laparoscopic prostatectomy procedure. By demonstrating the feasibility of using soft tissue navigation in complex surgical guidance procedures, we provide the next step in image guided interventions. In combination with hybrid tracers, such a navigation set-up may enable preoperative surgical planning as well as interventional navigation towards the lesions using the radioactive signature, while accurate target localisation and visualisation can take place using the fluorescent signature.

WP6: The public internet site that clearly explains the consortium, the goals of the project, and the technology that is being developed is available under

Project results:

New SiPM arrays for simultaneous time-of-flight PET/MR

The second part of the project was devoted to the improvement of the sensor performance in terms of energy resolution and time-of-flight capability and to the production of improved devices to populate an optimised pre-clinical system and a demonstrator of the clinical system. Concerning the improvement of the sensor performance, we worked both on the current n-on-p technology as well as on a new p-on-n version. By technology optimisation and the design of SiPMs with larger cells and thus a higher fill factor, a special single cell signal shape could be obtained. The main features of this signal are the very high fast component (having a duration < 10 ns) and the low recovery. These properties lead to a faster rising edge and a reduced baseline fluctuation when coupled to a LYSO crystal.

As a consequence, we find exceptional results in terms of energy and timing resolution with LYSO scintillator crystals. The key numbers obtained with this technology were presented at the IEEE Nuclear Science Symposium in 2010, Knoxville. We obtain below 11 % in a wide range of bias conditions. The CRT is below 325 ps. A simple leading edge discriminator (LED) was used in this timing measurement.

We also worked on signal filtering to further enhance the characteristics of the single-cell signal shown before. In particular, we started the development of a differential leading edge discriminator (DLED), in which we apply the LED to a signal obtained from the difference between the SiPM signal and a delayed replica. Thanks to the very slow recharge of the SiPM we are able to effectively compensate for the baseline fluctuation given by the pile-up of dark events. First CRT measurements are included in a paper submitted for publication on IEEE TNS. With a 5 mm short LYSO crystal, we reach about 230 ps FWHM CRT at room temperature, whilst decreasing the temperature to -30 oC we approach 180 ps. The values are slightly worse for a thick crystal because of the light propagation in the crystal itself: about 260 ps and 230 ps at 20 oC and -30 oC, respectively. These values are in line with the targets set at the start of the project. Using the technology described above we produced, tested and delivered about 450 fully working 2 x 2 arrays of 4 x 4 mm2 SiPM.

MR-compatible ASIC-tile

Foci in this task were the design of a new ASIC iteration PETA3, the design and production of a new detector stack 2.0 and the operation of the new stack with crystal arrays for functional verification.

Task 1.2.1: Performance Improvement

The improved TDC/ADC readout ASIC (PETA2) has been further characterised in detail. In particular, the influence of the various bias settings on noise, power consumption, energy- and time resolution has been studied. Typical noise values are in the range of 300 µV (rms) and below.

In order to better use the PETA chips for position interpolation with small crystals, it is very desirable to read also the small pulse amplitudes of edge channels. We have therefore added a neighbour logic to the ASIC which triggers the energy measurement of many channels when one large hit is seen. Further improvements in the PETA3 ASIC were an adaptation (reduction) of the analogue gain to the SiPM signal amplitudes (at high overvoltage), a slower baseline correction time constant, an on-chip current reference (to eliminate further components on the PCBs), and a successive approximation ADC with faster, constant conversion time of 1.33 µs.

PETA3 operates as expected and has been used in the new stack assemblies. The obtained flood map shows that most crystals of 1.37 x 1.37 mm2 can be clearly identified. The reconstruction algorithm used here is quite simple and further improvements are expected with more sophisticated methods.

Interfacing and packaging:

In order to operate the new PETA2 and PETA3 ASICs (with unavoidable slightly modified pin functionalities) and to eliminate all remaining magnetic components for MR compatibility, a new version of the detector stack (3 different PCBs) has been designed, manufactured and tested. As the standard gold plating layers of the PCBs (required here for wire bonding) contain (magnetic) nickel, it was decided to use an advanced nickel free ASIG (Autocatalytic Silver Immersion Gold) technology. Test PCBs were prototyped and sucessfully bonded by the assembly company. The manufacturing of the very dense PETA PCBs in this technology required several iterations until the bond traces had satisfactory width. Unfortunately, the actual batch of PETA PCBs turned then out to be too difficult to bond, so that a further round of (standard gold) PCBs has be manufactured and assembled. This caused a significant delay in the stack production. By the end of the project, all components for a full PETA3 based ring are available.

Time-of-flight MR-compatible PET module

The preclinical modules consist of a maximum of 6 detector stacks which are mounted on the singles processing unit board (SPU). Both components (stack and SPU) are made MR-compatible. Fluid cooling has been used to cool the stacks to about 20 °C operation temperature. Air cooling was added to avoid condensation inside the housing.

A preclinical scanner using SiPM based detectors with local digitisation:

The main components of the preclinical PET/MR scanner are a dedicated RF coil and the MR-compliant PET detector ring for imaging animals up to the size of rabbits. The system is designed to operate in a human clinical 3T MRI system (Philips 3.0T Achieva MRI).


The RF coil is a two channel transmit/receive coil in a birdcage design tuned to the proton (H1) frequency at 3T (f0=127.7MHz). 16 rungs have been chosen to allow homogeneous MR images even close to the inner cover of the RF coil. The bore size of the RF coil is 158 mm. The axial field of view (FOV) of the RF coil is 120 mm (3 dB drop of B1 on z-axis).

The main advantage of this stack is that all 8x8 analog signals of the silicon photomultipliers (SiPMs) are directly digitised by a customised ASIC board with 2x40 channels, which limits the possibility for cross-talk between the MRI and the PET detector electronics. Liquid and low level of air cooling were used to adjust the operating temperature of the detector modules to around 22 oC.

A stack consists of an LYSO array with 22x22 crystals of 1.3x1.3x10 mm3 coupled to an 8x8 channel SiPM array and an ASIC board with 2x40 channels. Each ASIC channel has a programmable threshold, which programs the leading edge discriminator (LED). If the threshold is passed by the input signal, the TDC latches the status of a PLL and an integrator with programmable integration windows is started. If the energy passes a programmable energy threshold, an ADC samples the integrated signal and signals a valid hit to the FPGA board. Time and energy resolution of this ASIC are 50 ps and 12 Bit.

All SiPM tiles were characterised in coincidence to a standard photomultiplier on a bench system to measure the flood histograms, energy histograms, and the mean light distributions for all 22x22 pixels.

The normalised mean light distributions of the SiPM tiles are used as model light distributions in a maximum likelihood positioning estimation (MLPE) algorithm. The algorithm computes the most likely crystal index in an iterative manner in which the scintillation process took place. The main reason to use an MLPE instead of a centre of gravity calculation was to compensate for missing channels which were caused due to trigger threshold variations in the currently used ASIC. The bias voltages of the SiPMs were adjusted to compensate for the gain variation of each SiPM tile. The ASIC parameters were adjusted to allow the detection of low light levels which lead to a sufficient sampling of the scintillation light distribution. The offset of each ASIC channel was measured with an atomised calibration procedure. During a measurement, PET detector raw data of all individual SiPM/ASIC channels were stored via a data acquisition computer on a fast disk. Typical storage size of such an acquisition is about 100 GBytes. Singles computation and coincidence processing was performed in post processing steps using the previously mentioned MLPE algorithm and sliding window techniques. Before the MLPE, the ASIC offset correction and a simple SiPM gain correction based on the shape of the ADC channel histograms were applied.

Interference investigation

The B0 disturbance has been investigated using a scaled phase measurement. A variation from black to white indicates a variation of from -1ppm to 1ppm, while a transition from white to black indicates a ppm (phase) jump. The image indicates a variation of about ± 4 ppm in the B0 field which is close to the specification of clinical MR systems.

The influence of the analog / digital electronics of the PET detector was investigated by measuring the RF receive noise of the MR system with the integrated RF coil during a measurement of a point source (22 Na, 2.4 MBq). Five noise measurements with identical bandwidths of f = 184.32 kHz and centre frequencies of f = f0+n*f (n = [-2,..2], f0 = 127.7 MHz) have been measured over a period of t = 40 s each. No disturbance of the RF noise was observed, as the noise pattern is identical to the one without the PET electronics inside the MR bore.

Finally, a simultaneous PET/MR image was measured using a hot-rod phantom with rod diameters of 4.8/4.0/3.2/2.4/1.6/1.2mm filled with 18F-FDG (9MBq) placed at the centre of the PET and MRI system. Acquisition time of PET was 22min. During PET acquisition, a b-TFE sequence (TE/TR: 1.8/4.9ms FA=50°) was used to acquire MRI data for 10min. For the singles and coincidence post processing, a singles cluster and coincidence window of 25 ns was used due to a currently not corrected systematic time-spread of up to 32 ns (no time alignment applied). Furthermore, neither normalisation nor calibration was applied. A ML-EM with 16 iterations was used to reconstruct the PET images. Image fusion was done with manual registration.
The PET image shows that rods with a diameter of 1.6 mm could be resolved. The large background in the PET image is expected to be reduced as soon as calibration, normalisation and time alignment are applied.

As part of the SUBLIMA FP7 project, we have developed a deformable elastic phantom that is both MR and PET visible - this allowed a complete test of synchronised simultaneous PET-MR acquisition which was successfully performed by imaging this phantom as it was periodically compressed and released by an MR-compatible stepper motor system.

Various diagnostic software tools were developed to assist with setting up and troubleshooting the system.

A tool to rapidly create and display sinograms was created - this allowed rapid identification of issues associated with for example, timing misalignments between modules etc. Many simple software bugs could be readily resolved as soon as they were identified.

A simple but fast SSRB-FBP reconstruction in the STIR environment was written. The full iterative PSF reconstruction provides superior image quality, however for initial setting up, troubleshooting and as a basic quality control procedure it proved very useful to have a way of obtaining preliminary images soon after acquisition, even if not of the optimum quality.

Calibration and normalisation

As soon as ready access to radioactive phantoms was available it was possible to perform a first pass optimisation of the basic parameters of the system including timing window and timing alignment, gain optimisation (given the limitations of the PETA1 ASIC), energy calibration and crystal identification algorithms.

A simple sinogram-based fan beam normalisation procedure was implemented to account for random and systematic variations in crystal efficiency. Normalisation data was acquired from a purpose-designed cylindrical phantom.

Investigations towards simultaneous clinical time-of-flight PET/MR

A split gradient coil and a gamma-transparent integrated transmit/receive RF coil have been designed and manufactured. The right image shows how up to 8 MR-compatible PET modules can be mounted inside the PET gantry.

Both coils (split gradient coil and RF coil) and two PET modules have been mounted in a 3T MR system. The straight elements are made from of a very light supporting material, a multi layer PCB on top to guide the RF current for the B1 field. Initially this RF body coil was designed as an eight channel transmit receive coil. However, for ease of operation, the coil architecture was converted into a 2 channel RF Transmit/receive coil, allowing operation in a standard 3T MRI system. The red, orange, transparent, and black wires/tubes are the communication, power and cooling wire/tubes for the PET detector, which are located in the top part of the split gradient coil.

Electromagnetic simulations are used to optimise the Tx/Rx coil design, which has to include the necessary shielding for the PET detector: modules, parts of the PET backbone and mounting structure. For the preclinical systems, one Tx/Rx coil for mice and one for rabbits will be designed. They can be used initially for the first test setup and later for fully equipped preclinical setups. A TEM resonator is suitable for the shielding approach. Accompanying simulations will serve to investigate the influence of the coil on the PET detector (scatter and attenuation of gamma quanta), to help to select suitable materials and geometries for the MR coil. A whole-body test setup is designed and built with a diameter of the patient port guaranteed to be >50 cm. Therefore, the quadrature body coil (QBC) and the gradient system will be modified to give sufficient room for the PET system, without degradation of the MR image quality in the field-of-view of interest. The RF properties of the modified RF coil as well as the gradient performance are simulated in detail before realisation.

Status of the clinical gradient and RF coil:

The whole body RF coil for simultaneous PET/MR imaging has been designed and is now in production. The coil is a Multi-element transmit/receive coil with 8 independent elements. All lumped components are moved out of the PET FoV in order to reduce the scattering of the high energy PET photons. Therefore a PCB based technology using distributed capacitors has been used to manufacture these individual stripes. The entire design allows a PET ring with an axial length of 20 cm.

The gamma transparent WB RF coil and the split gradient coil were mounted inside a 3T MR system within Philips to perform the interference tests with two PET modules.

Crosstalk minimisation

The first test setup for concurrent PET and MR data acquisition will already incorporate all mayor decoupling issues. Three main issues will be addressed to reduce crosstalk:

1. Minimisation of the inhomogeneity of the static magnetic field caused by the PET detector insert.
2. Reduction of the effective thickness and the area of ground layers and shielding, and thus reduction of eddy currents.
3. Exploitation of the Faraday principle to completely shield the PET electronics from RF emission pulses.

The minimisation of the crosstalk is addressed by simulations, correct use of components, specific shielding designs (transparent for gradient and resonant for RF) and building test detectors. Depending on the first experimental results obtained with the initial setup, further improvements of the decoupling will be obtained. Dedicated changes of the PET modules affect all partners of WP1. This can lead to changes in the design and might need iterative improvements. Selected test sequences of the MR will be used to identify the noise sources.

Interference test using the modified whole body PET MR system were executed in the last months of the project. As described above a split gradient coil which could cover up to eight ToF PET-modules was placed into a test bay of Philips into an 3T MR system. After mounting of the PET modules, the gamma transparent RF coil was placed inside the boar that the overall MR system functioned as a regular 3T MRI system. Feeding, cooling and communication of the modules were done from the backside of the MRI system. Several interference tests were executed. The PET modules were equipped with clinical LYSO crystals (4 x 4 x 10 mm^2).

Initially an RF leakage was found in the RF housing which leads to drop in the coincidence rate of two modules. The coincidence count rate drops to noise floor (randoms rate) when starting the MR sequence after 180 s due to insufficient shielding. Setup was modified afterwards to remove MR disturbances. The reason was a not well shielded RF cable that connects the backbone and the power supplies with the PET-modules. After the shielding was repaired, the coincidence count rate remains constant over time. A slight drop is still visible when starting the 1 kW RF pulses at 120 s.

However, when looking to the SNR with the image quality test tool provided the MRI system, a slight drop was observed in the SNR when the PET was switch on: PET OFF: SNR 205 versus PET ON (shielding repaired): SNR 185. However, no significant disturbance of the MR noise was observed during an RF noise tests when compared with PET on and PET off. The decrease in the SNR indicates that the noise floor is slightly increased during PET-on, which needs further investigations.

Finally, energy resolution and coincidence resolution time was measure under simultaneous MR operation (EPI: SE, TE/TR 60 ms / 2555 ms, FA = 90°). The coincident events were sorted by crystal number and the hottest 64 coincident pairs were used for calculating the average values for gain, energy resolution and timing resolution.

Under EPI acquisition, a timing (CRT) of 449 ps (FWHM) was measured which is to our knowledge the best ToF CRT that has ever reported under simultaneous PET/MR. Also, during this acquisition the singles and coincidence rate remains constant for all applied MR scans. PET gains and energy resolution remains constant during all different scans. The timing resolution varies due to long warm up phase after reset. Changes in timing are not related to MR scanning.

MR acquisition for motion modelling and concurrent PET/MR

We have considered in depth the problem of accurately measuring motion in the chest with MRI in order that this can be used to correct for the effects of motion in simultaneously acquired PET data. Our overall approach is to acquire a rapid dynamic 3D MR acquisition of the chest over a period of about 2 minutes and to use this to construct a model of the chest motion. During a subsequent PET-MR acquisition an MR 'navigator' signal is acquired - this parameter is able to drive the motion model and hence specify the complete chest anatomy as a function of time. The accuracy with which the navigator can predict the chest motion depends on several variables including in particular the complexity of the navigator. A 2D image navigator method has been developed and evaluated. This promises to fully specify both intra- and inter- cycle respiratory motion whilst taking only a small fraction of the total available MR acquisition time.

Further development of fast, robust, hierarchical, local-affine, non-rigid registration algorithm

- Fast, robust registration algorithm is essential for construction of respiratory motion models
- Three novel adaptive splitting techniques were evaluated - image-based; similarity-based; motion-based
- Regions of similar motion and/or image structure processed together to improve algorithm performance
- Evaluation on free-breathing whole-chest 3D MRI data from 10 volunteers and two publicly available CT datasets demonstrated <49 % reduction in registration error c.f. non-adaptive technique

Motion model driven by 1D MRI image navigator

- Subject-specific respiratory motion model constructed from near real-time dynamic MR images
- Model describes complex freeform deformations present in the human thorax during respiration.
- Real-time motion estimates based on a one-dimensional MR navigator
- Navigator is very fast and so may be interleaved with additional MR sequences
- Ability to predict motion evaluated using realistic PET simulation - clear improvement in visualisation of the myocardium and three tumours

Evaluation of optimum MR-navigator positioning

- Motion models based on single and pairs of navigators formed from volunteer MRI data
- The results suggest that two navigator models are more accurate, but only in the presence of significant cycle to cycle breathing motion variability.
- Overall the best individual navigator was positioned on the upper chest, and the best pairing consisted of navigators placed on the upper chest and lower chest.

Motion model driven by 2D MRI image navigator

- Free-form non-rigid respiratory motion correction in the thorax based on principal component analysis of the motion states encountered during different breathing patterns.
- 2D image navigator provides enough information to drive model and to differentiate between different breathing patterns.
- Estimates applicability of current motion model and hence need to update the model.
- Validated with MR data from 10 volunteers - demonstrated improvements of up to 40.5 % over other reported motion modeling approaches (i.e. including the combinations of 1D navigators above).

Motion compensated reconstruction

An environment has been developed in which a wide variety of different approaches to image-compensated reconstruction can be readily implemented and run on real and simulated datasets. Some of the more straightforward algorithms have been implemented and these all demonstrate improved image quality when applied to simulated data and data acquired from the pre-clinical MR-compatible PET systems that we have access to. It appears that regularisation is very effective in suppressing noise.

PET-MR simulation package

- Developed package for accurate simulation of simultaneously acquired dynamic PET data using anatomic and dynamic information from real MR acquisitions.
- Essential tool for development and evaluation of motion-compensated PET reconstruction techniques.
- Validated against standard 'GATE' Monte Carlo simulation.
- 150x faster than GATE for ten respiratory positions and 7000x faster for multiple realisations.
- Percentage difference of mean values is 3.1 % for tissue; 17 % for lungs; 18 % for small lesion.
- Demonstrated by simulating realistic PET-MR datasets from multiple volunteers with different breathing patterns.
- Will be distributed as part of the open-source STIR package, and example datasets will be made available to the community.

Implementation of motion-compensated reconstruction algorithms

- Tools for the development and evaluation of motion-compensated reconstruction have been implemented within the 'STIR' reconstruction package to allow maximum flexibility.
- Algorithms implemented include: reconstruct-transform-average (RTA), where gated are reconstructed individually and then transformed to a reference position and motion compensated image reconstruction (MCIR), where motion fields are incorporated directly into the system matrix.
- All reconstructions are based on the OSEM algorithm and variants of this including the ordered subsets maximum a posteriori one step late (OSMAPOSL) algorithm.

Demonstration of motion compensated reconstruction

- Motion compensated reconstruction has been demonstrated on various simulated and real data sets.
- Motion correction of simulated PET-MR data of the human chest.
- Motion correction for a locally rigid rotating phantom has been demonstrated using the 'PANDA' MR-compatible PET scanner - this demonstration has been extended to apply to simultaneous PET-MR data acquired in real-time.
- Motion correction of simultaneously acquired PET and MR data of a deformable PVA-cryogel phantom acquired with the HYPERIMAGE pre-clinical PET scanner / 3T-MRI, using a computer-controlled MR-compatible stepper motor arrangement.
- In all cases, motion-compensated PET reconstruction demonstrates improvement over non-corrected PET images.

Regularisation of motion compensated reconstruction algorithm

- The MCIR algorithm demonstrates significantly improved quantification and resolution properties compared to the simpler RTA algorithm, however MCIR also demonstrates significantly increased noise.
- We have demonstrated that the use of regularisation (e.g. median root prior) effectively suppresses noise.

3D and 4D attenuation correction derived from MR

The work on sequences and segmentation algorithms described in the first report has been extended to further define the accuracy required of MR-derived attenuation and to transform 3D MR-derived attenuation maps into a compete 4D attenuation map that can be incorporated into a motion correction scheme.

Effect of bone value assignment and segmentation on pet attenuation correction accuracy

- Evaluation of errors in attenuation corrected PET images resulting from inaccurate bone attenuation coefficient values and bone segmentation.
- Segmentation of CT components of clinical PET/CT images with mu-values assigned.
- Over-estimated bone volume with soft tissue mu-values produced the lowest error, with up to 36 %, 20 % and 10 % error (average maximum error over all patients) in the bone, soft tissue and lung, respectively.
- Using an over-estimated bone volume consisting of soft tissue equivalent attenuation coefficients presents a simple, robust method which is less sensitive to segmentation errors.

Development and evaluation of 4D motion compensated attenuation correction

- 3D and 4D attenuation corrections were evaluated on motion compensated 18F-FDG PET scans of several lung cancer patients acquired for radiation therapy planning.
- Attenuation and motion compensation were based on a respiratory correlated CT scan acquired just prior to the PET scan.
- No significant impact of 4D attenuation correction was observed regarding the tumour SUVmax / SUVmean, with, deviations up to 10 % for individual voxels

Requirements for motion model update frequencies

- Evaluated with repetitive time resolved Cone Beam CT based data sets of 32 lung cancer patients
- Variability of the motion amplitude at the lower parts of the lung was about 3 mm (1SD)
- From this data, a single motion model should suffice to motion compensate respiratory motion for periods in the order of four minutes.

Image reconstruction for preclinical system

- Iterative PET reconstructions for preclinical system developed and implemented.
- Reconstruction accounting for system geometry and gaps between modules.
- Reconstruction including motion correction options.
- FBP reconstruction for system assessment and troubleshooting.
- Normalisation and other calibration procedures based on uniform cylinder acquisition.

Applying MR-derived motion fields to MR-derived PET attenuation maps

- STIR simulation package used to evaluate impact of respiratory motion on MR-based AC.
- 4D attenuation map created by combining a respiratory gated UTE acquisition with a subject-specific motion model derived from a short dynamic MR acquisition.

Evaluation of the effect of errors in MR-derived attenuation map on reconstructed PET images

- Simulation study using the digital XCAT phantom indicates that five different tissue types need to be discriminated: air; soft tissue; lung; cortical bone; spongeous bone.
- Equivalent study with digital rat phantom (ROBY) shows only air, soft tissue and lung need to be discriminated.

Novel combination of MR sequences to obtain attenuation map

- MR sequence for human head / neck examinations Implemented for use on Achieva 3T MR
- Ultrashort-echo time (UTE) signal sampling for bone localisation combined with multiple gradient echoes for separation of soft tissue and adipose tissue.
- Combination of both in a single-shot sequence is a novel approach, targeting higher accuracy of attenuation map and lower scan times.

Data analysis and visualisation

Pharmacokinetic analysis of dynamic-contrast-enhanced MRI and dynamic PET requires measuring the concentration of the contrast-agent / tracer as function of time with high temporal resolution. Specific DCE-MRI sequences have been setup (with sampling rate up to 1 Hz) and tools for conversion of DCE-MRI raw data to contrast agent (CA) concentration have been developed. A program to fit the concentration of CA in tissue with a variety of PK models has been developed. The program allows simple visualisation of the results. In figure b, the overlay of the registered PK analysis result (perfusion constant) on a co-registered CT scan is shown and in figure c the FDG-PET (SUV) is overlaid on the same CT scan. Finally, in figure d and example of a PK-fit to the dynamic signal of a region of interest.

Substantial heterogeneity in tumour characteristics has been demonstrated. Clinical studies should reveal the relevance of this heterogeneity and test the hypothesis that radiation therapy dose redistribution tailored to this heterogeneity improves tumour control.

Combined sequential PET/CT and MRI examination was performed to show the benefit of concurrent PET/MR examination. To this end a number of parallel PET/CT and MRI data sets are available for longitudinal tumour growth and response monitoring. Unfortunately the techniques that were studied at the NKI-AVL have not been able to predict response better than 'conventional' tumour size measurements: 18F-FDG, 99mTc-Annexin V, CE MRI, and MR-Spectroscopy. This data has been supplemented with the use of targeted radiotracers for the somostatin receptor, the chemokine receptor 4 and alpha-V beta-3 integrins.

Registration of MR features to quantifiable values observed in PET, thus improving the early detection and staging of breast cancer using concurrent imaging

Sixty five patients with stage II/III breast cancer scheduled to be treated with NAC were investigated with MRI and PET. Scans were acquired prior and during treatment. At histopathology 32 patients showed complete response after NAC and 33 residual disease. The area under the ROC curve was 0.86 for MRI alone, and 0.94 in combination with PET, yielding 84 % sensitivity at 95 % specificity. The accuracy was largest for triple-negative tumours at both imaging modalities, but MRI and PET were complementary in the ER+/HER2- and in the HER2+ subgroups.

In other 32 patients with breast cancer dedicated hanging breast imaging was obtained using a high-resolution PET ring scanner (MAMMI PET). In 31 patients the primary tumour was visualised. In one patient the primary tumour was occult on MAMMI PET as well as on conventional PET/CT and MRI. Agreement of tumour FDG uptake between PET/CT and MAMMI PET was high (r = 0,86). However SUVmax was almost 2.5 times higher with the MAMMI PET compared to conventional PET/CT. Dedicated MAMMI PET was able to detect small lesions and could show heterogeneous tumour uptake, which may be important for FDG-guided biopsy in the future (see below section on image guidance).

Image navigation

In extension of the phantom efforts described under task 5.1.3, we also applied the developed navigation set-up in a patient. Following our previous experiences in hybrid surgical guidance during robot assisted procedures, we initially focused our clinical proof of concept on laparoscopic navigation towards the tracer deposits in the prostate. In spite of the challenging logistics and robotic arms blocking the detection of the reference targets by the optical tracker system at certain times, it was possible to use the navigation system during a robot assisted laparoscopic prostatectomy procedure.

To maximise the intraoperative navigation accuracy, a relatively rigid location (spina iliaca anterior posterior) near to the location where the fluorescence endoscope is inserted was found to be most suitable for the placement of the reference target. During surgery, the navigation system enabled accurate navigation to the prostate and was able to provide real-time distance estimations of the tip of the fluorescence endoscope to the centre of radioactivity / fluorescence within the prostate. Navigation of the tracked fluorescence endoscope towards the target identified on SPECT/CT resulted in real-time gradual visualisation of the fluorescent signal in the prostate, thus providing an intraoperative confirmation of the navigation accuracy.

By demonstrating the feasibility of using soft tissue navigation in complex surgical guidance procedures, we provide the next step in image guided interventions. In combination with hybrid tracers, such a navigation set-up may enable preoperative surgical planning as well as interventional navigation towards the lesions using the radioactive signature, while accurate target localisation and visualisation can take place using the fluorescent signature. Evidently the step from SPECT to PET merely depends on the isotope used and in phantom studies we have already demonstrated that such an approach can be expanded with MRI. Similarly the technology may easily be expanded towards the use of biopsy needles, rather than fluorescence laparoscopes.

Potential impact:

Socio-economic impact and the wider societal implications

The two leading causes of death worldwide are cardio vascular diseases (CVD) and cancer. Each year CVD causes 4.35 million deaths in Europe, mainly via coronary heart disease (CHD) and stroke. In 2004, 2 million Europeans (25 countries) were diagnosed with cancer (54 % men, 46 % women), and 1.2 million died of cancer (56 % men, 44 % women) [1, 2]. Intrinsically, disorder-initiated therapies allow for treatments only in a very late stage of a disease, resulting in low cure rate, high discomfort and much higher cost. Therefore, early detection of diseases and therapy response monitoring are the key requirements towards a more successful, less painful, and cost efficient treatment. Biomedical imaging and especially molecular imaging (MI) plays a key role. Here, PET has emerged as one of the MI techniques of choice, primarily due to its capability to detect tiny amounts of biomarkers. This quality was boosted in 2006 with the introduction of time-of-flight (ToF) PET, providing even higher effective sensitivity. Clinical studies have shown that ToF leads to improved lesion detestability, more homogeneous image quality, and more accurate quantification [3].

However, PET as a stand-alone modality has disadvantages: it suffers from a lack of anatomical detail, a strong dependence of the image quality on the patient size, a long acquisition time, and from blurring artefacts caused by respiratory, cardiac and patient motion, severely limiting a reliable quantification of the acquired data. To compensate for the limitations of PET, CT was successfully added to provide supplementary anatomical information to the excellent molecular image. However, it is known that this sequential hybrid imaging approach shows significant artefacts due to patient motion, particularly respiratory and cardiac motion, e.g. for cardiac PET. In contrast to CT, MRI offers superior soft tissue contrast and a wide range of functional information like quantitative flow, perfusion, diffusion, and blood-oxygenation - during the same acquisition. Additionally, dynamic MRI offers continuous measurement of organ motion using massive parallel MRI and does not burden the patient with ionised radiation.

One of the achieved goals of HYPERIMAGE was to develop a new MR-compatible detector technology with local digitisation. The main advantage of this local digitisation is scalability. The scalability allows the identification of very small crystals due to a multichannel detector arrays and the possibility to measure the time-of-arrival information with a high accuracy. Due to this, the HYPERIMAGE technology is the enabler for simultaneous ToF-PET/MR. To our knowledge this project demonstrated ToF-PET with sub nanosecond-time-resolution within a 3T MRI for the first time. Furthermore, the project clearly shows the feasibility to integrate this technology into a true simultaneous hybrid modality. Thus, it is to be expected that this novel hybrid modality will influence the biomedical research and the industrial communities significantly, and the consortium of HYPERIMAGE has stimulate European research in the area of hybrid imaging.

With the leading partners of various areas (from silicon detector technology over advancing methods to preclinical and clinical research), the HYPERIMAGE project extended the European research in this and adjacent areas to a leading position of Europe. On the long range it is the aim to improve the European healthcare system, by enabling novel investigations, and monitoring and treatment of diseases. The unique combination of the HYPERIMAGE research partners will significantly contribute to the development of a hybrid imaging modality that will positively influence the non-invasive prediction, diagnosis, monitoring and prognosis of diseases. In the long run, new standards will be developed that support the planning and provide tools for the guiding of therapeutic interventions.

Contribution to non-invasive prediction of diseases

Besides the diagnosing of clinically apparent diseases and monitoring of therapy PET-MRI is assumed to play a major role in the individually tailored prediction of disease manifestation. Screening for known risk factors of cardiovascular diseases or pre-invasive stages of cancer is of major interest to identify people at risk. They could then be assigned to dedicated and cost-effective prevention programs or treatment regimens prior to clinical manifestation of the disease. The PET-MRI system exhibits favourable advantages for whole body screening applications. The combined assessment of morphological and molecular information is likely to enhance the specificity to confirm or rule out early pathophysiological changes associated with a later onset of diseases, such as vulnerable atheroma plaques leading to acute cardiovascular events. Recent investigations from V. Fuster's team have demonstrated the association of 18FDG PET in the atheroma plaque with known markers of plaque vulnerability. PET-MRI assessment of vulnerable atheroma plaque is going to have a high prevention and clinical cardiovascular impact in population with both moderate and high cardiovascular risk, selected by the current clinical scores for cardiovascular risk.

Furthermore, the low radiation exposure associated with PET-MRI minimises the potential harm of the screening examinations compared to the use of X-rays. In addition, 'one-stop-shop' protocols combing both imaging procedures into a short and single examination will improve patient comfort, thus leading to a far-reaching acceptance of this screening procedure by people.

Contribution to diagnosis of diseases

The increase in sensitivity and in quantification will improve the quality of diagnoses due to several reasons: for application in the area of oncology the detection of small lesions is heavily influenced by breathing and cardiac motion. Here, the concurrent image acquisition of molecular and functional data combined with motion compensation will allow precise and thus early diagnosis of cancer, see WP5 for details. Recent developments [Tobias / Paul] using model-based motion compensation of the lung by observing the movement of the arterial tree indicate the use of the PET/MR system for the detection of small lesions in the lung. Applications, like the assessment of the myocardial ischemia, require the investigation of the heart with MRI and PET under stress and rest. In the myocardial infarction, concurrent PET-MRI will allow a more precise assessment of myocardial metabolism because the best capability of motion compensation and a simultaneous assessment of wall motion and myocardial scar. Concurrent hybrid imaging will allow imaging of these two conditions in one session and will therefore not suffer from their non-reproducibility. By adding the concurrent measurement of the MR to the PET information physicians will have a novel tool allowing them to look fully synchronous at dynamic functional data and at dynamic biochemical processes. Depending on the diseases the physician will be able to adapt the contrast mechanism to focus her/his diagnosis on the very specific character of these diseases.

Contribution to monitoring of diseases

Compared to PET/CT, the aimed concurrent high sensitive PET/MR will allow very low doses imaging via reducing the amount of PET-tracer. These will enable periodical assessments of the current status of a disease including the response to pharmaceuticals, surgery, and other novel techniques like high intensity focus ultrasound (HiFU). For instance, it is known, that some tumours are resistant to particular treatment regimens, while they are highly sensitive to others. Accurate monitoring of the response to chemo-, radiation or other therapies is not only beneficial for the identification of tumour regression, but primarily for the adaptation of therapy in the case of non-responders. Additionally, motion correction will enable the adoption of novel tracer with higher decay times.

Contribution to prognosis of diseases

The prognosis of a disease that can be assigned to an individual patient depends on a variety of parameters, including the accuracy of diagnosis and the suitability of the treatment.

PET-MRI is a technology aimed to improve these two parameters. An increase in the accuracy of diagnosis by using concurrent PET-MRI information, such as tissue metabolism, molecular markers and morphological data, aids to better assign the patient to a certain disease stage which is required to apply more specific treatments. Furthermore, different treatment options are often available, and PET-MRI has the potential to screen for reliable predictors indicating a response on specific treatments.

Contribution to evaluate therapies, develop new ones (such as new medicines, and gene and cell therapies) and support

The PET-MRI system provides distinct possibilities to evaluate e.g. cancer therapy. The response to loco-regional and systemic cancer treatment by chemo- , immuno- and radiotherapy can be assessed upon first cycles of treatment by determination of early changes in tumour metabolism, perfusion and morphology. This will significantly improve the tailoring of established treatment regimen by adapting therapeutic and adjuvant pharmaceutical doses. Such individualised cancer treatments are likely to increase therapeutic outcome and, moreover, to minimise adverse side effects. Furthermore, these features can be used to evaluate new and alternative therapies, by rapid screening of large patient cohorts. Thus PET-MRI can assist in the developmental process of new pharmaceuticals during the clinical phase.

Contribution to plan and guide therapeutic interventions

The combination of PET and MRI offers unique features for diagnostic and therapeutic image guided interventions. MRI guided biopsies on tumour tissue for example can be targeted to specific areas within the tumours that reveal specific metabolic characteristics or immuno-receptor status as determined by PET. Regarding the planning of radiotherapy the supplemental information of PET-MRI can be used to more precisely target on viable tumour tissue in order to reduce radiation expose and side effects in patients. Moreover, MR guided therapeutic interventions such as thermo- and cryoablation as well as high intensity focus ultrasound would also benefit from more precise targeting. Thus, the improvement in safety and efficacy in image guided interventions will gain them an increased acceptance by performing clinicians and patients.

European approach

The HYPERIMAGE project compromises multiple scientific and technical challenges in various areas. These challenges are along almost every component of the novel hybrid imaging system. To ensure the accomplishment of the overall aim of the project all scientific areas have to be stuffed with leading research groups in their fields. More precisely, the project requires the leadership in high speed low noise and low power ASIC design, academic leadership in MR, academic leadership in PET, academic leadership in medical software and of cause the leadership in medical preclinical and clinical research. Especially, the development of the new detector technology and the high sensitive PET/MR system requires the involvement of leading industrial partners in the area of silicon detector technology and of PET and MRI system.

All leader groups are located in several European countries. There is no European nation that has all necessary leadership groups for carrying out this project on its own. Main advantage of this leading European team is the broad cross-linking to other research groups which stimulates adjoining and further research and development activities in these areas. It will further lead to an extension of the leading position of the EU in new hybrid imaging technologies.

[1] Petersen S., Peto V., Rayner M., Leal J., Luengo-Fernandez R. and Gray A. (2005) European cardiovascular disease statistics. BHF: London.
[2] WHO, 'Noncommunicable diseases in the WHO European Region: the challenge, Fact Sheet EURO/06/04.
[3] The benefit of time-of-flight in PET imaging: Experimental and clinical results, J.S. Karp et al., J Nucl Med 49(3): 462-70, Mar 2008

Public website:

Contatc details:

Project coordinator: Volkmar Schulz
Tel: +49-241-6003319
Fax: +49-241-6003442

Related information

Reported by

37, High Tech Campus (WY 3.045)
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top